Quantifying cardiac sympathetic denervation: first studies of 18F-fluorohydroxyphenethylguanidines in cardiomyopathy patients

4-18F-Fluoro-m-hydroxyphenethylguanidine (18F-4F-MHPG) and 3-18F-fluoro-p-hydroxyphenethylguanidine (18F-3F-PHPG) were developed for quantifying regional cardiac sympathetic nerve density using tracer kinetic analysis. The aim of this study was to evaluate their performance in cardiomyopathy patients. Eight cardiomyopathy patients were scanned with 18F-4F-MHPG and 18F-3F-PHPG. Also, regional resting perfusion was assessed with 13N-ammonia. 18F-4F-MHPG and 18F-3F-PHPG kinetics were analyzed using the Patlak graphical method to obtain Patlak slopes Kp (mL/min/g) as measures of regional nerve density. Patlak slope polar maps were used to evaluate the pattern and extent of cardiac denervation. For comparison, “retention index” (RI) values (mL blood/min/mL tissue) were also calculated and used to assess denervation. Perfusion polar maps were used to estimate the extent of hypoperfusion. Patlak analysis of 18F-4F-MHPG and 18F-3F-PHPG kinetics was successful in all subjects, demonstrating the robustness of this approach in cardiomyopathy patients. Substantial regional denervation was observed in all subjects, ranging from 25 to 74% of the left ventricle. Denervation zones were equal to or larger than the size of corresponding areas of hypoperfusion. The two tracers provided comparable metrics of regional nerve density and the extent of left ventricular denervation. 18F-4F-MHPG exhibited faster liver clearance than 18F-3F-PHPG, reducing spillover from the liver into the inferior wall. 18F-4F-MHPG was also metabolized more consistently in plasma, which may allow application of population-averaged metabolite corrections. The advantages of 18F-4F-MHPG (more rapid liver clearance, more consistent metabolism in plasma) make it the better imaging agent to carry forward into future clinical studies in patients with cardiomyopathy. Trial registration: Registered at the ClinicalTrials.gov website (NCT02669563). URL:https://clinicaltrials.gov/ct2/show/NCT02669563


Introduction
Sudden cardiac death (SCD) from a precipitous onset of cardiac arrhythmia is a common cause of death in patients with cardiomyopathy [1]. Increasing use of implantable cardioverter defibrillators (ICDs) and cardiac resynchronization therapy (CRT) devices as preventative therapy against SCD has reduced mortality rates in this population [2]. The primary measure used to select patients for ICD therapy is a left ventricular ejection fraction (LVEF) below 35% [3]. Unfortunately, LVEF alone is a poor predictor of who will benefit from ICD therapy, as only about 1 in 8 people who receive an ICD will experience a lifesaving shock [4]. Also, This article is part of the Topical Collection on Cardiology.
inappropriate shocks for arrhythmic events that are not lifethreatening are not uncommon and can be fatal [5].
Remodeling of cardiac sympathetic innervation occurs in many diseases and has been linked to an increased risk of ventricular arrhythmias, the most frequent arrhythmia to cause SCD [6]. Myocardial infarction causes sympathetic denervation zones which are often larger than the corresponding myocardial scar, demonstrating the sensitivity of sympathetic neurons to ischemic insult [7]. Alterations in ion channel function in denervated myocardium is one mechanism that can promote malignant ventricular arrhythmias [8]. The link between sympathetic remodeling and arrhythmogenesis has led to investigation of noninvasive cardiac denervation metrics, obtained using sympathetic innervation radiotracers, as predictors of SCD risk. The ADMIRE-HF trial with 123 I-meta-iodobenzylguanidine ( 123 I-mIBG) established the independent power of the heartto-mediastinum ratio (H/M ratio) measured with planar scintigraphy, a global measure of cardiac denervation, to predict SCD risk [9]. The PAREPET trial showed that the extent of regional cardiac denervation, measured with 11 C-(-)-m-hydroxyephedrine ( 11 C-HED) and PET, classified ischemic cardiomyopathy patients into tertiles with low, medium and high risk of SCD [10]. Also, several clinical studies have shown that higher H/M ratios for 123 I-mIBG and lower denervation extent metrics measured with 11 C-HED are linked to improvements in myocardial function and sympathovagal balance in cardiomyopathy patients treated with CRT devices [11][12][13]. These findings point to a potential role of cardiac neuroimaging in improved staging of patients for CRT or ICD therapy.
We recently developed two new sympathetic innervation radiotracers, 4-18 F-fluoro-m-hydroxphenethylguanidine ( 18 F-4F-MHPG) and 3-18 F-fluoro-p-hydroxphenethylguanidine ( 18 F-3F-PHPG). In addition to the advantages of their fluorine-18 radiolabel, these tracers were designed to have improved kinetics for more accurate quantification of regional nerve density using tracer kinetic analysis techniques. They have slower neuronal uptake rates than 123 I-mIBG and 11 C-HED and much longer neuronal retention times (i.e., they have "irreversible" tissue kinetics) [14]. Slowing neuronal uptake rates reduces flow-limitation effects that occur for tracers with rapid uptake into tissue compartments, which in principle should greatly improve sensitivity for detecting early denervation [15]. Irreversible tissue kinetics affords high tracer tissue concentrations for imaging despite the slower neuronal uptake rates and allows for the application of alternative kinetic analysis techniques. Specifically, studies with 18 F-4F-MHPG and 18 F-3F-PHPG in non-human primates and healthy human subjects showed that the Patlak graphical method [16] provides reproducible metrics of regional nerve density [14,17]. The goal of this study was to perform a head-to-head comparison of 18 F-4F-MHPG and 18 F-3F-PHPG in cardiomyopathy patients selected for ICD placement to determine which has the best overall properties for measuring cardiac sympathetic denervation using PET.

Image analysis
After image reorientation, short-axis images were interpolated into 4-mm slices (PMOD v3.8, PMOD Technologies). The left ventricular wall on the final frame was divided into 60 sectors using an algorithm written in IDL (v6.2, Harris Geospatial Solutions). Using 18 short-axis slices encompassing the left ventricle, 18 × 60 = 1080 volumes-of-interest (VOIs) were defined, and tissue time-activity curves C t (t) were extracted for each VOI. A whole-blood time-activity curve C wb (t) was obtained from a VOI placed in the basal left ventricular blood pool.

3
18 F-3F-PHPG. Radio-HPLC analysis of purified plasma aliquots (100-300 μL) determined the fraction of plasma activity associated with intact radiotracer, f intact (t). The f intact (t) vs. time data were fit to dose-response functions using Graph-Pad Prism v3.03 as previously described [18]. Whole-blood and plasma aliquots (100 μL) were counted in a gamma counter to measure the plasma to whole-blood activity concentration ratio, C p /C wb . This tended to be constant during a study, so the mean ratio C p ∕C wb was calculated. A plasma time-activity curve, C p (t), was estimated from the wholeblood curve as:

Tracer kinetic analysis
Kinetic analyses were performed using PMOD's Kinetic Modeling (PKIN) module. For 18 F-4F-MHPG and 18 F-3F-PHPG, Patlak plots were constructed from the tissue and plasma time-activity curves and analyzed with linear regression to generate regional Patlak slopes, K p (mL/min/g), and the linear regression coefficient r, as previously described [17]. Regression analysis used data from the 4.5 min frame to the final frame. 13 N-Ammonia kinetics were analyzed using the DeGrado method [19] to obtain regional perfusion estimates, F (mL/min/g). Using polar maps of the Patlak slopes or perfusion estimates, the fraction of VOIs with nerve density or perfusion estimates less than a cutoff threshold of 50% of the maximum map value were calculated to estimate the extent of left ventricular denervation or hypoperfusion.

Tracer retention measures
For comparison with Patlak analysis, tracer "retention index" (RI) metrics for 18 F-4F-MHPG and 18 F-3F-PHPG were calculated by dividing the final tracer tissue concentration C t in each VOI by the integral of the whole-blood time-activity curve, C wb (t), using the Tracer Retention algorithm in the PKIN module of PMOD. The fraction of VOIs with RI values less than a cutoff threshold of 50% of the maximum RI was calculated to estimate the extent of cardiac denervation.

Safety and tolerability tests
Safety tests of 18 F-4F-MHPG and 18 F-3F-PHPG were performed before and after each PET session, including a resting 12-lead ECG, and measurements of heart rate, blood pressure, respiration, and body temperature. Blood tests (comprehensive metabolic panel, complete blood count, plasma catecholamine levels) and urinalysis were also performed. Heart rate and peripheral capillary oxygen saturation (SpO 2 ) were continuously monitored during PET scanning, and blood pressure measured every 10 min. Subjects were contacted at 24 h and 30 days to enquire about any adverse events.

Statistical analysis
Statistical calculations were performed using Microsoft Excel 2016. Mean tissue activity concentration ratios for peak heart-to-liver (A), peak heart-to-blood (B), and peak heart-to-lung (C) in the last five PET image frames

Subject characteristics
Clinical data for the 8 subjects are summarized in Table 1. Six subjects had ischemic cardiomyopathy with prior myocardial infarction. The other two had nonischemic cardiomyopathy with no known infarctions. None had mixed etiology. Single-chamber ICD devices were most often implanted, while one was a dual-chamber ICD, and another was a biventricular ICD with a cardiac resynchronization therapy defibrillator (CRT-D). Concomitant medications included the statin atorvastatin (n = 6); the selective serotonin receptor inhibitor (SSRI) citalopram (n = 1); the mixed nonselective beta-blocker and α 1adrenoreceptor antagonist carvedilol (n = 4) or the β 1adrenoreceptor antagonist metoprolol (n = 3); the angiotensin II type 1 receptor (AT 1 ) antagonist losartan (n = 2); and the angiotensin-converting enzyme (ACE) inhibitors lisinopril (n = 5) or ramipril (n = 1).

PET Imaging
Seven of the eight subjects completed all three PET scans. One subject (#1) ended their innervation scans early (at t = 40 min) due to back pain. Another subject (#3) left the study after completing scans with 13 N-ammonia and 18 F-4F-MHPG. Representative fused PET/CT scans are shown in Fig. 1. 18 F-4F-MHPG and 18 F-3F-PHPG produced high-quality cardiac PET images, with low lung uptake and good heart-to-blood contrast. More prolonged retention of 18 F-3F-PHPG in the liver relative to 18 F-4F-MHPG is evident in the coronal images.

Tissue concentration ratios
Peak heart uptake levels and mean concentrations in blood, liver, and lung were determined in the last five image frames to calculate the tissue ratios: peak heart-to-liver, peak heartto-blood, and peak heart-to-lung (Fig. 2). 18 18 F-3F-PHPG also had consistently higher peak heart-tolung ratios (Fig. 2C), averaging 40% higher than those of 18 F-4F-MHPG (1.4 ± 0.0). This was significant from 17.5 to 45 min (p < 0.007 to 0.02), but not at 55 min (p < 0.09).

Blood sample analysis
Metabolism of 18 F-4F-MHPG and 18 F-3F-PHPG in plasma was biphasic (Fig. 3). A large fraction of 18 F-4F-MHPG is rapidly metabolized in the early phase, with times of 50%

Tracer kinetic analysis
Patlak analysis of 18 F-4F-MHPG and 18 F-3F-PHPG kinetics was successful in all regions for all subjects. Figure 4B shows examples of 18 F-3F-PHPG kinetics in regions covering a range of nerve densities. Corresponding Patlak plots are shown in Fig. 4C. Highly linear fits of the Patlak plot data were obtained, independent of the degree of denervation. Linear correlation coefficients r (mean ± SD within a subject) ranged from 0.965 ± 0.025 to 0.995 ± 0.003 for 18 F-4F-MHPG and from 0.990 ± 0.009 to 0.995 ± 0.005 for 18 F-3F-PHPG. Patlak slopes and resting perfusion estimates are presented in Table 2. The distribution of Patlak slopes in each subject is illustrated in Fig. 5. All subjects had some regions with Patlak slopes in the normal range, representing regions with preserved innervation, as well as regions with Patlak slopes below normal, consistent with partial or complete denervation. Within subjects, regional Patlak slopes for 18 F-3F-PHPG and 18 F-4F-MHPG were highly correlated, with linear correlation coefficients r ranging from 0.911 to 0.985 (data not shown). Estimates of the extent of left ventricular hypoperfusion and sympathetic denervation from threshold analysis of the polar map data are provided in Table 3. Polar map examples are shown in Fig. 6. Denervation extent measures for 18 F-4F-MHPG and 18 F-3F-PHPG closely agreed in most subjects, and differed by < 5% for all subjects. Correlations between regional Patlak slopes and resting perfusion estimates varied considerably across subjects, with R 2 values ranging from 0.123 up to 0.728 (Fig. 7).

Tracer retention analysis
Retention index (RI) values and denervation extent estimates obtained using RI values are shown in Table 4. Within individuals, there was strong linear correlation between regional RI values and Patlak slopes, with linear correlation coefficients r > 0.991 in all cases for both tracers. This is expected since areas with higher Patlak slopes will generate higher amounts of tracer trapped in neurons. However, the slopes of the correlations were variable across subjects, ranging widely from 0.41 to 0.89 for 18 18 F-3F-PHPG is also tied to its slower metabolism. Finally, denervation extent metrics from RI values were in close agreement with those obtained from Patlak slopes (Fig. 8A, B). Strong linear correlations between the two denervation extent metrics were seen for both tracers (Fig. 8C), with R 2 = 0.996 (p < 2 × 10 -8 ) for 18 F-4F-MHPG and R 2 = 0.995 (p < 5 × 10 -7 ) for 18 F-3F-PHPG.

Discussion
First-in-human studies of 18 F-4F-MHPG and 18 F-3F-PHPG in healthy subjects demonstrated that their irreversible kinetics can be analyzed with Patlak analysis to estimate regional cardiac sympathetic nerve density, with Patlak slopes averaging 0.107 ± 0.010 mL/min/g and 0.116 ± 0.010 mL/min/g, respectively [17]. The first studies of these tracers in cardiomyopathy patients reported here show they can also provide nerve density measures in hearts with extensive disease-induced denervation. Their administration caused no changes in blood pressure, heart rate, ECG measures, or laboratory tests and caused no adverse events, further establishing their safety. Comparing 18 F-4F-MHPG and 18 F-3F-PHPG, each agent demonstrated an advantage over the other. 18 F-4F-MHPG had much higher heart-to-liver ratios due to its faster clearance from the liver, providing better heart-tobackground contrast. More importantly, the faster liver clearance of 18 F-4F-MHPG reduces spillover of liver counts into adjacent ventricular regions, which can confound Patlak analysis in those regions ( Figs. 1 and 2). 18 F-3F-PHPG had consistently higher heart-to-blood ratios, due to moderately lower blood concentrations and slower metabolism in plasma which increased myocardial concentrations. 18 F-3F-PHPG metabolism in plasma was not only slower than 18 F-4F-MHPG, it was also much more variable (Fig. 3). The cause of this is unknown, but could be related to medication effects in the liver. A potential advantage of the more consistent metabolism of 18 F-4F-MHPG is that it may allow the application of population-averaged metabolite corrections for generating input functions, eliminating the need for blood sampling. In kinetic analyses, regional Patlak slopes obtained with 18 F-4F-MHPG and 18 F-3F-PHPG were highly correlated in all subjects and there was also very good agreement between each tracer's estimate of the extent of left ventricular denervation (Table 3). Thus, the two tracers are essentially equivalent in their ability to quantify regional nerve densities. 18 F-4F-MHPG and 18 F-3F-PHPG accumulate in cardiac sympathetic nerve varicosities as substrates of the norepinephrine transporter (NET) and are rapidly stored in norepinephrine storage vesicles by the second isoform of the vesicular monoamine transporter (VMAT2). Efficient retention in storage vesicles is the mechanism responsible for their irreversible tissue kinetics [15]. Our results show that 18 F-4F-MHPG and 18 F-3F-PHPG retain their irreversible kinetics in patients with cardiomyopathy,  Fig. 7 Scatter plots of regional Patlak slopes vs. regional resting perfusion estimates, showing correlation coefficients (R 2 ). Cutoff thresholds set at 50% of the maximum Patlak slope or perfusion estimate are shown as dashed lines. A 18 F-4F-MHPG data for Subject #8. B 18 F-4F-MHPG data for Subject #3. C 18 F-3F-PHPG data for Subject #7 including cardiac regions with severe nerve losses. This is an important finding since vesicular storage function is an energy-dependent process that can be compromised during acute ischemia [20]. The observation that 18 F-4F-MHPG and 18 F-3F-PHPG maintain their irreversible kinetics in regions with substantial nerve losses indicates that vesicular storage function remains intact in the surviving neurons of patients with chronic cardiomyopathy. All subjects had regions with Patlak slopes in the normal range (Fig. 5) and the myocardial kinetics of the tracers were irreversible, evidence that concomitant medications did not interfere with tracer uptake and storage. Statins like atorvastatin may reduce the elevated sympathetic outflow seen in cardiomyopathy and can restore sympathovagal balance [21]. Thus, statins may reduce extraneuronal norepinephrine concentrations via reduced outflow. The SSRI citalopram has a binding affinity of only 4.1 mM for NET [22]. At typical blood levels of 0.1-0.7 mg/L [23], equal to concentrations of 0.3 μM-2.1 μM, citalopram should not interfere with cardiac NET transport. The beta-blockers carvedilol and metoprolol do not possess intrinsic sympathomimetic activity and primarily act to inhibit post-synaptic responses to norepinephrine [24]. Also, they do not cause large changes in sympathetic outflow [25], so they should not inhibit NET transport. Supporting this, a meta-analysis of the ADMIRE-HF trial data found that H/M ratios for 123 I-MIBG were independent of beta-blocker dose and plasma norepinephrine levels [26]. The renin-angiotensin-aldosterone system is activated as a compensatory mechanism in the failing heart, increasing production of angiotensin II by angiotensin-converting enzyme (ACE). Angiotensin II binding to vascular AT 1 receptors potently induces vasoconstriction. It also acts on presynaptic AT 1 receptors on sympathetic nerves to facilitate norepinephrine release, which contributes to vasoconstriction through vascular α 1 -adrenoreceptors [27]. AT 1 inhibitors like losartan inhibit the pressor effects of angiotensin II and also reduce enhanced norepinephrine release. Similarly, ACE inhibitors like lisinopril and ramipril lower angiotensin II levels, reducing sympathetic outflow and increasing vagal tone [28]. In either case, AT 1 inhibitors and ACE inhibitors each act to reduce sympathetic outflow and thus are unlikely to interfere with tracer uptake into neurons.
Most of our subjects were treated with ICDs, but one had a CRT device implanted. While ICDs react only during arrhythmic events, CRT devices are continuously active, providing biventricular pacing to resynchronize ventricular contractions in patients with serious rhythm disturbances. Studies with 123 I-mIBG and 11 C-HED have shown that more highly preserved sympathetic innervation at baseline predicts better responsiveness to CRT [12,13]. Also, in CRT-responsive patients, 123 I-mIBG exhibits higher late H/M ratios and slower washout rates than at baseline, along with better autonomic balance, consistent with improved sympathetic function [11]. The one CRT-treated patient in our study had VOIs with Patlak slopes in the normal range at baseline, indicating intact tracer uptake mechanisms in those regions. However, it is possible that nerves in other regions with below normal Patlak slopes could have reduced neuronal function that would improve in response to CRT. Future studies with 18 F-4F-MHPG or 18 F-3F-PHPG in CRT patients could potentially identify the specific regions that functionally improve, providing insights into CRT-induced changes in presynaptic NET and VMAT2 function.
An advantage of Patlak analysis is it only requires linear regression of the transformed kinetic data, a robust method that can be applied to small region sizes. In preclinical studies in nonhuman primates with 18 F-4F-MHPG, Patlak slopes showed good reproducibility (± 10% or less) under control conditions [14]. Also, in pharmacological blocking studies using different doses of the potent NET inhibitor desipramine (DMI) to establish varying fractions of available cardiac NET transporters, Patlak slopes declined with increasing DMI doses following a sigmoidal dose-response model, showing that they sensitively tracked declines in available NET. At the highest dose of DMI (1.0 mg/kg), blood activity exceeded myocardial activity, consistent with The side-by-side presentation of perfusion and innervation polar maps (Fig. 6) and the plots of regional innervation metrics vs. perfusion estimates (Fig. 7) illustrate how nuclear cardiologists could compare the pattern and extent of sympathetic denervation with regional perfusion to assess SCD risk. The observed patterns between regional nerve density and perfusion depend on the complex history of the progression of each subject's cardiovascular disease and treatments. For example, in some subjects, the polar maps showed that revascularization had restored perfusion to areas that remain denervated (e.g., Fig. 6A and B). Zones of denervated myocardium that are well perfused ("perfusion-innervation mismatch" zones) are areas that promote the genesis of ventricular arrhythmias [29]. In this study, five of the eight subjects had denervation zones that were larger than the area of hypoperfusion (Table 3). All five had ischemic cardiomyopathy with prior infarctions. Conversely, two subjects with matched hypoperfusion and denervation zones had nonischemic cardiomyopathy and no known infarctions. Whether this pattern difference between ischemic and nonischemic cardiomyopathies would be a consistent finding in a larger cohort remains to be seen, since the PAREPET trial excluded patients with nonischemic cardiomyopathy. However, PAREPET evaluated a perfusioninnervation mismatch metric which demonstrated power in predicting arrhythmic risk, but the denervation extent measure alone was found to be a stronger predictor of sudden cardiac arrest [10].
Clinical studies with 123 I-mIBG and 11 C-HED rely on semi-quantitative metrics of tracer retention, such as the H/M ratio for 123 I-mIBG or RI values for 11 C-HED, as measures of global or regional sympathetic nerve density, respectively [7,9]. While these measures have served as useful first approaches to quantifying nerve losses, a fluorine-18 PET radiotracer that can accurately and reproducibly quantify regional nerve density using kinetic analysis methods is highly desirable, not only for routine clinical studies but also for research on the impact of diseases or new therapeutics on cardiac nerve populations [30]. In this study, we compared nerve density and denervation extent measures obtained using Patlak slopes and RI values. While regional RI values were highly correlated with regional Patlak slopes within individual subjects, the slopes of these correlations varied, due to variation in tracer metabolism rates. Because individual tracer metabolism rates are unaccounted for in the RI calculation, the variation in RI values across individuals limits their utility as nerve density metrics on an absolute scale. However, denervation extent estimates from RI values were within a few percent of those obtained from the Patlak slope data (Fig. 8). Therefore, if the sole purpose of a clinical PET study with 18 F-4F-MHPG or 18 F-3F-PHPG is to estimate the extent of cardiac denervation, RI data would be adequate for this purpose and would avoid blood sample analysis. On the other hand, if the goal was to obtain quantitative measures of regional nerve density in addition to the denervation extent measure, Patlak analysis would be the method of choice.  Limitations of this study include the small number of subjects and the absence of female subjects in the study cohort, which limits its generalizability to all cases of cardiomyopathy. Future studies will include a similar head-to-head study design comparing the performance of 18 F-4F-MHPG with 11 C-HED for quantifying the extent of left ventricular denervation.

Conclusion
In conclusion, 18 F-4F-MHPG and 18 F-3F-PHPG can each be used to quantify sympathetic nerve density at high regional resolution using Patlak analysis, a unique attribute that distinguishes them from existing cardiac innervation tracers. Nerve density and denervation extent measures obtained with the two agents were comparable. The strengths of 18 F-4F-MHPG include its more rapid clearance from the liver, reducing spillover from the liver into the inferior wall, and its more consistent metabolism in plasma, which may make it possible to avoid blood sampling by using a population-averaged metabolite correction curve. These two advantages outweigh the higher heart-to-blood contrast of 18 F-3F-PHPG, making 18 F-4F-PHPG the better agent to carry forward for further clinical development. Our results support the performance of larger clinical trials to establish the power of nerve density and denervation extent metrics obtained with 18 F-4F-MHPG to predict clinical outcomes in patients being evaluated for CRT or ICD therapy.